Landslide susceptibility mapping using hybrid random forest with GeoDetector and RFE for factor optimization
Citation
Zhou, X., Wen, H., Zhang, Y., Xu, J., Zhang, W. (2021). Landslide susceptibility mapping using hybrid random forest with GeoDetector and RFE for factor optimization. Geoscience Frontiers, 12(5): 101211. Link to paper
Abstract
This study develops two hybrid feature-optimization workflows (GeoDetector-RF and RFE-RF) to improve landslide susceptibility mapping performance over a baseline random forest model. A landslide inventory with 406 landslides and 2030 non-landslide samples was used together with 22 conditioning factors, and optimization was performed to reduce redundancy and collinearity before model training. The optimized-factor models were further analyzed with frequency ratio and multicollinearity diagnostics, then evaluated using accuracy and ROC-AUC metrics. Both hybrid models outperformed the baseline RF model, indicating that factor optimization can improve model robustness and predictive reliability. The work provides a practical data-driven strategy for selecting influential factors in susceptibility assessments.